Vision Enhanced and Noise Destroyed Generative Adversarial Network for Improving Deep Learning based Tomato Leaf Diseases Prediction

Read full text Summarise with VIRAT
Authors: Palle Swetha, N. Sharmila Banu
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Early disease prediction in tomato plants enables early diagnosis and treatments, reducing crop damages and promoting healthier yields.Deep Learning (DL) models are effective tools for predicting tomato plant diseases at an early stage.Several models utilized Generative Adversarial Networks (GAN) to generate noise free images for efficient prediction.The combination of CycleGAN and the Faster Region-based Convolutional Neural Network (FRCNN) were utilized for disease localization and prediction.However, CycleGAN faces challenges with slow convergence and low-resolution image output.Hence, in this paper, the Vision Enhanced and Noise Destroy GAN (VEND-GAN) model is proposed to solve the above-mentioned issues for predicting tomato leaf diseases effectively.VEND-GAN employs two generators, Generator 1 for noise reduction and Generator 2 for resolution enhancement.Generator 1 utilizes the Mutual Attention Mechanism (MAM) to model complex noise distributions.This mechanism models the intricate noise patterns in images, improving the precise renewal of the actual image content.Generator 2 applies Residual in Residual Dense Block (RRDB) for enhancing the image resolution that captures the fine details and textures of tomato plants.RRDB enhances the gradient flow of image features captures complex patterns and simplifies training by focusing on input-output differences, improving feature representation across multiple scales.The generated Noise Reduced and High Resolution (NRHR) images are evaluated by a single discriminator that distinguishes fake from real images.The discriminator model is combined with the Principal Component Analysis and Deep Neural Networks (PCA-Deepnet) feature extraction module to obtain feature loss.PCA-Deepnet extracts features of original and NRHR images to calculate feature loss.The joint loss function which includes adversarial loss and feature loss, preserves essential features and structural similarity in generated images.Thus, VEND-GAN enhanced images, retaining the most discriminative features and helps for improving and classification accuracy.The NRHR images are directly fed into FRCNN for disease localization and prediction.The complete framework is termed the Deep Tomato Disease Detection network (DeepTDDnet).Finally, the DeepTDDnet model achieves 97.27 % and 97.24% accuracy, surpassing hybrid Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) and hybrid Convolutional Neural Network -Vision Transformer (CNN-ViT) models for the PlantVillage Dataset and the Kaggle Tomato Leaf Diseases Dataset, respectively.

Fields of Study & Subjects

Smart Agriculture and AI Advanced Neural Network Applications Image Enhancement Techniques

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
654–669
Publication date
23 March 2026
Language
EN